Rewrite Resume

Rewrite Resume

Tailor the resume to one target role. Interview first, write second — every strengthened claim has to be true.

Output location

Everything for one target role lives in one folder:

$JOB_SEARCH_DIR/<company>-<role-slug>/
├── inputs/                        # raw job posts, the resume you provided
├── keyword-report-2026-08-24.md   # every output is dated; newest wins
├── linkedin-rewrite-2026-08-24.md
└── resume-rewrite-2026-08-24.md

JOB_SEARCH_DIR is the environment variable if the user has set one, otherwise ~/job-search. Resolve it once with Bash — echo "${JOB_SEARCH_DIR:-$HOME/job-search}" — and never hardcode a path. Create the folder on first run and tell the user where it went.

The newest report is whichever keyword-report-*.md sorts last: ls "$DIR"/keyword-report-*.md | tail -1.

This skill writes resume-rewrite-<YYYY-MM-DD>.md. Dated by run, never overwritten — a second run the same day gets a -2 suffix, so the folder keeps every version you've sent. Provided resumes and pasted JDs go in that folder's inputs/.

Step 1 — Get the keyword report

Check $JOB_SEARCH_DIR for an existing <slug>/keyword-report-*.md — reports are dated; use the most recent. If one matches the target role, read it and confirm the target with the user.

No report? Ask for the job link(s) and mine them first:

• Fetch chain (never ask for LinkedIn credentials): logged-in Chrome → WebFetch → ask the user to paste the JD. For Chrome, load tools with ToolSearch query select:mcp__claude-in-chrome__tabs_context_mcp,mcp__claude-in-chrome__navigate,mcp__claude-in-chrome__get_page_text,mcp__claude-in-chrome__computer.
• Extract per job: title, seniority, years, hard skills, tools, soft skills, certs, exact recurring phrases, responsibilities, must-have vs nice-to-have.
• Aggregate into a keyword frequency table (keyword | count | category | where to showcase), top 10 keywords, and phrases to mirror verbatim.
• Save as keyword-report-<YYYY-MM-DD>.md in the role folder so the LinkedIn skill can reuse it.

Step 2 — Ingest the current resume

• Markdown / text — read it directly.
• PDF — use the Read tool (it handles PDFs with a page range).
• DOCX — use the built-in docx skill to extract the content.
• No resume — offer to build one from scratch through the interview in step 3. Say so plainly; don't silently invent a starting point.

Copy whatever they provide into inputs/ so the run is reproducible.

Step 3 — Interview loop (do not skip)

Ask before writing:

• Quantify — every achievement bullet that has no number: "How much, how many, how fast, compared to what?"
• Scope — team size, budget, users, revenue, systems owned
• Tools — for each tool the JD names: production use, dabbled, or never? Ask directly.
• Verify — for each claim you want to sharpen: "Is this accurate, and to what degree?"
• Missing wins — "What did you ship that never made it onto the resume?"
• Gaps and dates — employment gaps, overlapping roles, contract vs full-time. Get the facts; don't paper over them.

Batch with AskUserQuestion where answers are choosable; plain chat for numbers and stories.

Honesty rules — non-negotiable:

• Never invent experience, metrics, titles, dates, or employers.
• Every claim traces to the resume, provided context, or a user answer.
• Unverified suggestions are marked [CONFIRM] in the draft.
• Sharper wording, yes. Inflated facts, no.

Step 4 — Section-by-section rewrite

Each section as:

Current → what's there now Suggested → the rewrite Why → which report keywords it hits and what got stronger, in one plain line.

Cover:

1. Summary / headline — 2–3 lines, targeted at this role, leading with the strongest true claim.
2. Skills section — mirrored to the JD's own language (if they say "CI/CD pipelines", don't write "build automation"). Group by category, order by report frequency.
3. Experience bullets — action verb + what you did + measurable outcome + JD keyword. One idea per bullet. Cut bullets that serve no keyword and show no outcome.
4. Projects — keep the ones that prove a top-10 keyword; cut or shrink the rest.
5. Education & certifications — placement (top if recent grad, bottom otherwise), and which report-named certs are worth pursuing.

Beyond wording, give structural recommendations:

• Reorder — which sections move up for this specific role
• Cut — stale roles, obsolete tech, anything older than ~10–15 years that isn't load-bearing
• Add — missing sections the JD implies (publications, speaking, open source, languages)
• Length — one page under ~10 years experience, two pages beyond; say which applies and what to cut to get there

Step 5 — ATS pass

• Keyword coverage table — top report keywords vs where they now appear. Flag every miss and say whether it's fillable honestly.
• Formatting warnings — call out anything that breaks resume parsers: tables, multi-column layouts, text in headers/footers, graphics and icons, text inside images, non-standard section headings, unusual fonts. Recommend a standard single-column layout with conventional headings ("Experience", "Education", "Skills").

Step 6 — Save and report

Write the dated resume-rewrite-<YYYY-MM-DD>.md containing:

1. The clean final draft, ready to copy out
2. A change log — what moved, what was cut, what was added, and why

In chat print the new summary and the top 5 changes only. Point at the file for the rest.

Offer: "Want the LinkedIn profile matched to this? Run rewrite-linkedin-profile — it'll reuse the same report."